During the research on the theoretical framework of policy gradient reinforcement learning, it is proved that the gradient estimation formulas of all the existing policy gradient algorithms can be uniformed.

For the problem of multi-wheel coordination in motion control of lunar rover, an adaptive control method based on hybrid policy gradient reinforcement learning has been proposed.

针对月球车运动控制中的多轮协调问题，提出了一种基于混合策略梯度增强学习的自适应控制方法。

互联网摘选

A hybrid policy gradient reinforcement learning control method is proposed to solve this complex optimation control problem with difficulty in obtaining teacher signals and designing fuzzy rules.

针对这种导师信号难以获取、模糊规则难以制定的复杂优化控制问题，本文提出了一种基于混合式策略梯度增强学习PG-SVM的多轮协调控制方法。

互联网摘选

The fuzzy policy gradient reinforcement learning, which incorporates a priori knowledge by using fuzzy inference systems, has been studied in this dissertation.

研究了利用模糊推理系统引入先验知识的策略梯度增强学习算法。

互联网摘选

A method for optimal reward-baseline to minimize the variance of gradient estimation is presented and the method is proved theoretically.

提出了一种求解最优回报基线的方法，使得策略梯度估计的方差减小到最小。

互联网摘选

The Optimal Reward Baseline for Policy-Gradient Reinforcement Learning

策略梯度强化学习中的最优回报基线

互联网摘选

In this paper, we analyze some reinforcement learning methods, which are Value-based reinforcement learning ( VBRL), Policy-Gradient reinforcement learning and Actor-Critic reinforcement learning etc.

本文分析了几种强化学习方法，包括基于值函数（Value-Based）近似方法、策略梯度方法（Policy gradient）、以及Actor-Critic方法等。

互联网摘选

According to this framework, some current policy gradient algorithms are generalized. 2.

并且在上述理论框架的指导下，对现有的策略梯度算法进行了推广。

互联网摘选

Two fuzzy policy gradient reinforcement learning algorithms are proposed for Markov Decision Processes with discrete and continous actions, respectively.

本文分别针对具有离散行为空间和连续行为空间的马氏决策问题，提出了两种模糊策略梯度增强学习方法（Fuzzy Policy Gradient：FPG）。

互联网摘选

Currently, the main network security hot spot analysis system is based on natural language processing techniques of which the key analysis methods is retrieving key information from massive amounts of data based on theme-based model, LDA model and N-gram model.

当前主流的网络安全热点分析系统的研究和开发主要是基于自然语言处理技术，里面分析热点的关键方法是基于主题模型的LDA模型[4]和N-gram模型[7]从海量数据中提取重点信息。

互联网摘选

Now, chunk identification is widely used in many fields of natural language processing, especially in the example based machine translation ( EBMT), in which chunk identification is one of major techniques.

现在组块分析广泛用于自然语言处理的众多方面，尤其是在基于实例的机器翻译EBMT研究中，组块分析是重要技术之一。

互联网摘选

The operation document understanding is an important job in C3I System. Template-based method and natural-language-based method are most commonly used now, but these two methods are less flexible and cannot well express the semantic relationship in special fields.

作战文书的理解是C3I系统中的一项重要工作，目前普遍采用的是基于模板和基于自然语言处理两种方式，存在着适应性差、不能很好表示特定领域语义关系。

互联网摘选

Word Sense Disambiguation ( WSD) plays an important role in Natural Language Processing ( NLP). The study on WSD has great theoretical practical significance in Natural Language Understanding ( NLU) and now it has become a hotspot and nodus.

在自然语言处理（NLP）中，词义排歧（Word Sense Disambiguation）一直是研究的重点和难点，对其他的语言信息处理任务具有重要的理论和实践意义。

互联网摘选

Processing NIL text requires unconventional linguistic knowledge and techniques. Developed to handle formal language text, the existing natural language processing methods exhibit less effectiveness in dealing with NIL text.

NIL表达处理需要非常规知识和技术，而目前面向正规语言的自然语言处理技术在处理NIL文本时效果并不理想。

互联网摘选

According to the principle of use of Web information extraction can be divided into six kinds of way. For example, information extraction of based on wrapper, information extraction based on HTML structure, information extraction based on natural language processing, and so on.

根据采用的原理可以将Web信息抽取分为六种方式，例如基于包装器语言的信息抽取、基于HTML结构的信息抽取、基于自然语言处理方式的信息抽取等等。

互联网摘选

This article discusses the standardization stage of information processing& information content, which human beings have reached. Information content standardization depends on the integration and development of Knowledge Base construction and XML/ RDF for natural language processing.

本文论述了人类信息处理已经达到信息内容的标准化阶段，信息内容的标准化有赖于自然语言处理中的知识库建设和可扩展标记语言（XML）、资源描述框架（RDF）的融合与发展。

互联网摘选

The experiment makes clear that choosing domain concept as character item is better than others. Its F1 of macro-average is 79.35%, and its F1 of micro-average is 88.00%.

实验表明，在自然语言处理中，选择领域概念作为特征项，其宏平均下的F1值为79.35%，微平均下的F1值为88.00%。

互联网摘选

Two methods for collecting the structured data are discussed, which are natural language processing ( NLP) and structured data entry ( SDE).

阐述了两种结构化数据采集的方法：自然语言处理（NLP）和结构化数据输入（SDE）。

互联网摘选

These types of systems include advanced natural language processing systems that can discover new terms and relationships by analyzing content ( such as document text in ECM systems).

这种系统包括高级的自然语言处理系统，它能够通过分析内容（比如ECM系统中的文档文本）发现新的词汇和关系。

互联网摘选

Lots of advanced AI techniques such as machine learning, knowledge representation, pattern identification, genetic algorithms and distributed intelligent systems have been applied to the research ofDSS.

许多先进的人工智能技术如机器学习、知识表示、自然语言处理、模式识别、遗传算法及分布式智能系统都被融入DSS的研究中。

互联网摘选

Co-Training, as an alternative to EM algorithm, is a well-known form of bootstrapping which is a topic of interest in NLP.

作为EM算法的替代，Co-Training是众所周知的自举算法，近来已经成为自然语言处理领域的兴趣焦点。

互联网摘选